Sunday, October 3, 2010

Is Stock Picking Really Dead? Part III

Have you ever watched a game of little kid soccer?  Little kids, not big kids.  Here's a hint from my town.  At this age, the moms are not yet soccer moms.  They still think they're hipsters from Brooklyn.  The dads won't show, out of fear someone will recognize them as a New Jersey resident.

Anyway, in Kiddie Soccer, the ball moves randomly.  The four year old kids move randomly near the ball, but not always.  Kid movements correlate poorly with each other and with the ball.  The average location of the kids tells you the ball's location, roughly.  It certainly tells little about the location of any particular kid.

By age six, the kids all chase the ball well.  The dense pack of furious kickers sticks closely to the ball!  The average location of the kids tells you the precise location of the ball.  Not only that, it tells you pretty well location of each and every kid!  Correlation with each other kid (and the ball) rises.  The ball, more or less, still moves randomly, but even without seeing it, you know where it is.

[Maybe by high school kid movements have lower correlation with each other: They stay in position on the field.  And, ball movements get predictable: running and passing in the right direction to reach the goal.]

Keep these images in mind.  They'll help in my third post on the death of stock picking.  In the first and the second, I explained that a tendency by stock pickers to buy smaller and cheaper (in a technical sense) companies helped their performance dramatically in the post internet bubble period, right up until the "Great Recession" began, and those tendencies have been a headwind ever since.  Hence, the systematic death of stock picking.

This post, about correlation, will also help explain why, in the original WSJ story, everyone seems to argue macro investment themes are the way to invest today.

This discussion is difficult.  Please bear with me.  You need to understand how correlation across stocks affects the chances that stock pickers can outperform, through luck or skill.

A stock index, like the S&P 500, is a market capitalization weighted average of a bunch of stocks, meant to reflect "the market".  Imagine the S&P 500 like the soccer pack.  The index could move up, with all companies in a tight pack around the average.  The index could move down with companies all behaving very differently, or any other combination in between.  Does this make sense? Let's hope so, because it's time for a...
Pop Quiz!  
Question 1: What must be true for the S&P500 to move down with all companies bouncing in different directions?  Answer: The biggest companies must fall!  They make up a greater share of the index.If the biggest companies go up, the chance that the index rises must, well, fall.

Question 2: If the S&P 500 moves up, could 480 of the stocks in it be falling?  Yes. Remember, the biggest companies have the greatest influence on results.  So, if the 20 biggest companies win a little, that could overwhelm 480 little losers.  What about an index of the same 500 names that weights them equally?  Still possible, but somewhat less likely.  The 20 winners would have to go up more than the 480 losers.
Okay, still with me?  Good.  Suppose 10 portfolio managers each pick 10 stocks for their portfolio randomly, and they'll hold 10% positions in each of them.  If all stocks are highly correlated (the tightly grouped pack of soccer players) then random picking doesn't lead to much variation in outcomes. All the managers will look pretty much the same after the fact.  And, they'll all likely look pretty much like the index.  The same holds if they have skilled picks instead of random picks.  A portfolio of the ten best stocks may not perform substantially better than the ten worst.

However, if correlations are very low, 10 random stocks picked by 10 managers could result in vastly different results.  Obviously, if a manager picks a top outlier, she has a 10% position in a huge performer.  Similarly, if bad luck strikes and she has a bottom outlier, her business has ended. 
But, just as importantly, outcomes will look very different if any of the outliers happen to be heavily weighted in the index.  
Take an extreme example: 498 stocks are flat.  One rises by 100%, one goes bankrupt, (falls by 100%.) First, assume these outliers are very small index components, say 0.01% of the index, (like New York Times Co and Eastman Kodak Co.)  If your portfolio has the winner, you're hailed as a genius because your portfolio is up 10%.  If your portfolio has the loser, you're going out of business.  If you had neither, you're muddling along like the index, and like everyone else.
Now assume the winning outlier contributes hugely to the index, like  Exxon Mobil, at over 3%.  If you didn't randomly (or "skillfully") draw the winner, you're out of business because you underperformed the index so badly.  Nothing else really matters.
What does this mean for the death of stock picking?  Correlation helps drive periods when stock picking looks good.  When correlations across stocks are very high, zero-skill punters on stocks can hide in the pack, and high skill pickers have a hard time differentiating themselves.  Similarly, when correlations are low, lucky managers with no skill can look pretty darn smart, while skilled stock selectors could be hurt by bad luck.

Now, back to the WSJ story: Is stock picking dead?  In Part I, we saw size biases made stock picking easy in the post bubble period and hurt in the most recent environment, independent of skill.  In Part II, we saw the well documented "value" approach to stock picking that usually helps actually hurt.  Finally, we see that in periods of high individual stock correlation the stock picking game gets even harder.

What's an investor to do??


  1. Terence the Incompetent Armchair TheoreticianOctober 5, 2010 at 6:56 PM

    How much work has been done on the difference between the model where prices move according to some random distribution as opposed to the fact that buying behaviour determines the price? The problem with your soccer ball analogy is that prices don't move randomly, they are determined by the flocking activity of the people chasing the ball. It seems like stock prices are really more like high school popularity (where essentially all of the "price" comes from the behavior of the participants) than soccer. I suppose I could be wrong in that price is moved by independent market participants that have more effect than your average mutual fund. Dunno.

    If you can model price as popularity, then it would seem that a collectively anxious market would result in higher correlations (as funds chase winners and shun losers), and that a market with sunnier participants would result in lower correlations as buyers were willing to buy current losers with the hope/knowledge that things are improving.

    My guess is that this kind of model blows up all kinds of underlying IID assumptions in standard models, but I was curious if there's been any attempt to build models of this type.

  2. @Terence:

    I've tried to respond several times...terrible technical good excuse!

    To continue to abuse the analogy, all that ever really matters is the unpredictable part of the ball movement. Once the high school players stay in position, and the defender kicks the pass to the open forward, everyone "knows" the ball is moving from the defender to her teammate. However, the only interesting question: can the now defending team cause a surprise in that path, or will the offensive team cause their own surprise by missing the pass? Realize, I'm only using the soccer analogy to highlight correlation issues, I'm not using it as a model of securities pricing.

    However, you raise the interesting question posed often by behavioral finance folks: Are there other things that drive prices? In terms of modeling, behavioral finance has a long way to go. However, good experimental evidence abounds.

    I'll defer to Robert Lucas for his "Defence of the Dismal Science" piece in The for an ungated version if you're interested...which explains why classical economic theory still has deep relevance in the post "Great Recession" world.

  3. It seems to me we could perform a historical survey of stocks in the index to determine each one's correlation coefficient with the performance of broader market over, say, the last year. Independently we also perform whatever subjective or objective rating system we like to predict a stock's prospects. Now we plot the various stocks in a plane with correlation coefficient as one axis and prospects on the other axis. By picking stocks in the happy quadrant you should be able to outperform the broader market during times of high correlation.
    Q1 "Happy": Moderate ("low") correlation, good prospects
    Q2 "Wishlist": High correlation, good prospects
    Q3 "Downer": Moderate ("low") correlation, bad prospects
    Q4 "Untouchable": High correlation, bad prospects

    Does it not follow, assuming we are correctly assessing the market as being overcorrelative?

  4. Mr. Freeze, or do you prefer to go by Carbon? I'm not exactly sure what you mean by the term "overcorrelative"? Maybe you mean "overly correlated"? In any case, your four quadrant analysis is a standard one for selecting "a low beta, cheap" portfolio. Low beta is a fancy (more precise) way to say low correlation and "cheap" means good prospects, presumably.

    However, neither of these measures says anything about the market being more or less correlated. I think the best way to think about this is that you are saying individual stocks are either high or low, not really addressing the overall spread. My point is that if the high correlations are at 95%-99% and the lows are at 85%-90% that is a very different environment than when the highs are at 75%-80% and the lows are at 50%-60%.